We develop a new technique, based on a low-energy theorem (LET) of NSVZ type derived in [1], for the nonperturbative investigation of SU(N) QCD with N f massless quarks -or, more generally, of massless QCD-like theories -in phases where the beta function, β(g), with g = g(µ) the renormalized gauge coupling, admits an isolated zero, g * , in the infrared (IR) or ultraviolet (UV). In the above phases, the theory is either exactly conformal invariant -if g(µ) = g * for any scale µ -or may be asymptotically conformal in the IR/UV -if g(µ) = g * for some µ = 0. In the latter case, g(µ) → g * only for µ → 0 + /µ → +∞. We point out that the LET sets constraints on 3-point correlators involving the insertion of Tr F 2 , its anomalous dimension γ F 2 , and the anomalous dimensions of multiplicatively renormalizable operators at g * . These constraints intertwine with the exact conformal scaling for g(µ) = g * and the IR/UV asymptotics -which may or may not coincide with the IR/UV limit of the aforementioned conformal scaling -for g(µ) = g * . Our new technical tool is the nonperturbative evaluation of the dimensionally regularized LETin the conformal-invariant scheme introduced in [2] -either exactly for g(µ) = g * or asymptotically in the IR/UV for g(µ) → g * . In the above cases, we also discuss how the LET for bare correlators is the rationale for the existence in massless QCD of the mysterious divergent contact term in the OPE of Tr F 2 with itself discovered in perturbation theory in [3,4] and computed to all orders in [5]. Specifically, if γ F 2 does not vanish, the divergent contact term in the rhs of the LET for the 2-point correlator of Tr F 2 has to match -and we verify by direct computation that it actually does -the divergence in the lhs due to the nontrivial anomalous dimension of Tr F 2 . Hence, remarkably, the additive renormalization due to the divergent contact term in the rhs is related by the LET to the multiplicative renormalization in the lhs, in such a way that a suitably renormalized version of the LET has no ambiguity for additive renormalization. E.3 LET for γ F 2 = 0 in d = 4 72 E.4 LET for exceptional γ F 2 = 0 in d = 4 73 E.5 LET for O = F 2 to order g 2 in perturbation theory 73